[Patch v2]: Fix ld pr11138 FAILures on mips*.

[Richard Sandiford]

David Daney <david.daney@cavium.com> writes:
> I will wait a couple of days to give Richard a chance to object.

Looks good to me too.  Thanks for doing this.  I think it should go
on the 2.22 branch as well.

> I am especially concerned about what happens on IRIX where the symbol
> comes in from an external object rather than being generated by the
> linker itself.  I had no way to test that.

Me neither, unfortunately.  But I agree it looks right.  There's a
possibility that we could create the dynamic sections before the
definition of __rld_obj_head has been read in (e.g. from the
check_relocs of a previous object).  In that case it looks like
we would create the .rld_map section and __rld_map/__RLD_MAP symbol,
then later switch to __rld_obj_head.  .rld_map would then be left
at zero size and the __rld_map/__RLD_MAP symbol would be ignored
when setting the tag value.  But your patch preserves the
behaviour in that case too.

Richard